|Publication number||US6571332 B1|
|Application number||US 09/546,979|
|Publication date||May 27, 2003|
|Filing date||Apr 11, 2000|
|Priority date||Apr 11, 2000|
|Also published as||DE60006007D1, DE60006007T2, EP1272937A1, EP1272937B1, WO2001077843A1|
|Publication number||09546979, 546979, US 6571332 B1, US 6571332B1, US-B1-6571332, US6571332 B1, US6571332B1|
|Inventors||Paul C. Miranda, Larry D. Hewitt, Stephen C. Ennis|
|Original Assignee||Advanced Micro Devices, Inc.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (7), Referenced by (27), Classifications (12), Legal Events (5)|
|External Links: USPTO, USPTO Assignment, Espacenet|
1. Field of the Invention
This invention relates to computer systems and, more particularly, to system memory transactions.
2. Description of the Related Art
In general, computer systems include a microprocessor, system memory, a video graphics adapter and a plurality of peripheral devices. The memory and peripheral devices are typically coupled to the microprocessor through one or more system buses. In personal computers (PCs), these buses are controlled by bridging logic and controllers which may be separated into different integrated circuits: The system controller and the peripheral bus controller. The system controller, commonly referred to as a northbridge in PC systems, includes bridging logic, a memory controller and an accelerated graphics port (AGP) controller and a peripheral bus controller.
Many peripheral devices and the graphics port commonly need access to the system memory through the memory controller. Typically, theses devices request memory accesses or transactions in a particular order, but receive the transaction responses in a different order. Therefore the various peripheral controllers and the graphics port controller must use some method of keeping track of which responses correspond to which requests. One method of keeping track of the requests and responses is to attach a linearly incrementing tag to each request. A request tag table is then used to store the request tags. As the responses come back, they can be sorted and reordered by matching the response tags with the request tags in the request tag table. The reordered responses can then be stored in a response buffer. Additional logic, such as a free buffer space counter, may be employed to prohibit the response buffer from being overflowed by not allowing requests to issue while the response buffer is full. Additional methods may employ a large enough response buffer that the free buffer space counter is not necessary.
Although the above methods may provide a method of reordering transactions and keeping track of which responses correspond to which requests, it is desirable to use a different method to provide buffer reordering and management.
The problems outlined above may in large part be solved by a method and apparatus for transaction reordering and buffer management. In one embodiment, the apparatus may include a buffer, a first generator circuit and a second generator circuit. The buffer is configured to store memory transaction responses received from a memory controller in a plurality of addressable locations. The first generator circuit is coupled to the memory controller and is configured to generate a first memory transaction request encoded with a first tag corresponding to an address in the buffer in response to receiving a first memory request. The second generator circuit is coupled to the first generator circuit and is configured to generate a second tag using the size of said first memory request added to the first tag.
In an additional embodiment, the first generator circuit may be further configured to generate a second memory transaction request encoded with the second tag corresponding to a second address in the buffer in response to receiving a second memory request successive to the first memory request. The second generator circuit may be further configured to generate a third tag using the size of the second memory request added to said second tag.
The generation of transaction requests encoded with tags which correspond to addresses in the transaction response buffer based on the size of the requested transaction may advantageously simplify the process of reordering memory transaction responses.
Other objects and advantages of the invention will become apparent upon reading the following detailed description and upon reference to the accompanying drawings in which:
FIG. 1 is a diagram of one embodiment of a computer system.
FIG. 2 is a block diagram of one embodiment of a transaction buffer management apparatus.
While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the drawings and detailed description thereto are not intended to limit the invention to the particular form disclosed, but on the contrary, the intention is to cover all modifications, equivalents and alternatives falling within the spirit and scope of the present invention as defined by the appended claims.
Referring to FIG. 1, a diagram of one embodiment of a computer system is shown. The computer system of FIG. 1 includes a microprocessor 100 coupled to a northbridge 110 through a system bus 105 and a southbridge 180 coupled to northbridge 110 through a peripheral bus 175. Northbridge 110 includes a memory controller 130 coupled to a system memory 155 through a memory bus. Northbridge 110 also includes a peripheral bus controller 140 coupled to a peripheral device 170 through peripheral bus 175 and an accelerated graphics port (AGP) controller 120 coupled to a graphics adapter 160 through a graphics bus 165. Both AGP controller 120 and peripheral bus controller 140 are coupled to memory controller 130 allowing memory transactions to occur. AGP controller 120 includes a transaction buffer management apparatus 200, which will be described in more detail below.
It is noted however, that in other embodiments, the location of the components included in northbridge 110 and southbridge 180 may be different. For example, AGP controller 120 may be physically located in southbridge 180 or it may be part of microprocessor 100. The above system architecture of FIG. 1 is shown only as an example of how a system may be configured.
In one embodiment, a controller such as AGP controller 120 may make a system memory request to memory controller 130. The request may be a read request or a write request. Logic in AGP controller 120 may separate read requests from write requests by using two different request queues. As will be described further below in FIG. 2, any request whether a read request or a write request, will be encoded with a tag. The tag is used to differentiate the requests from one another. Additionally, the requests may be sent in a particular order, but due to prioritization schemes, the responses may come back from memory controller 130 in a different order, thereby creating the need to keep track of the requests and their corresponding responses.
Turning now to FIG. 2, a block diagram of one embodiment of transaction buffer management apparatus 200 is shown. Transaction buffer management apparatus 200 includes a transaction response buffer 205 coupled to memory controller 130 of FIG. 1. Transaction response buffer 205 of FIG. 2 is also coupled to a buffer pointer 210. Buffer pointer 210 is coupled to a tag comparator 250. Tag comparator 250 is coupled to a transaction request generator circuit 220. Transaction request generator circuit 220 is coupled to a tag generator circuit 230 and a tag register 240.
In general, when a memory request 226 arrives at transaction request generator circuit 220, memory request 226 is encoded with a tag which is held in tag register 240. The tag is generated by tag generator circuit 230 and stored in tag register 240. Before a transaction request 225 is released to a transaction queue, tag comparator 250 compares a next tag 231 to a current response tag 211 pointed to by buffer pointer 210. A more detailed example follows below.
When a first memory request arrives at transaction request generator circuit 220, tag register 240 is initialized to #0h and represents the initial tag corresponding to the first location in transaction response buffer 205. The memory request is encoded with this initial tag. Tag generator circuit 230 determines the size of memory request 226 and generates a next tag 231 by adding the size of memory request 226 to the value of the initial tag. The tags generated by tag generator circuit 230 represent addresses in transaction response buffer 205. In this case, if memory request 226 were 8 bytes, then next tag 231 would be #8h.
Once next tag 231 is generated, tag comparator 250 performs a series of calculations to either enable or disable transaction request generator circuit 220 from releasing transaction request 225 to a transaction queue. The first calculation takes the difference between current response tag 211 and a current request tag 241 stored in tag, register 240 using binary unsigned subtraction. The second calculation takes the difference between next tag 231 and current request tag 241 using binary unsigned subtraction. Then tag comparator 250 compares the first calculation with the second calculation. If the first calculation is greater than the second calculation, then tag comparator 250 allows transaction request enable 255 to enable transaction request generator circuit 220 to release transaction request 225 to a transaction queue. When a second memory request 226 of 16 bytes, for example, is received by transaction request generator circuit 220, memory request 226 is encoded with the tag held in tag register 240, which is #8h. Tag generator circuit 230 generates next tag 231 by adding the 16 bytes to the tag held in tag register 240. In this example, next tag 231 would be #18h. Next tag 231 is stored in tag register 240 awaiting the next memory request. In order to enable transaction request generator circuit 220 when the first memory request arrives, a buffer empty bit may be used to override the result of the compare calculations performed by tag comparator 250, since current request tag 241 and current response tag 211 are both zero. In alternative embodiments, other features may be used to enable the first transaction request such as forcing tag register 240 to start at a non-zero value.
To prevent response buffer overflow, tag comparator 250 prevents transaction request generator circuit 220 from releasing transaction request 225 to the request queue if a response to transaction request 225 may overwrite data in buffer 205. When tag generator circuit 230 reaches the last addressable location, the next tag will wrap around to #0h. The above tag encoding process is repeated each time a memory request is received by transaction request generator circuit 220.
As transaction response 201 comes back from memory controller 130 of FIG. 1, it is stored in transaction response buffer 205 of FIG. 2. Each transaction response 201 contains data and a response tag which may be an address or a partial address in transaction response buffer 201. Transaction response 201 is stored at the address specified by its tag. Buffer pointer 210 points to the address of the transaction response 201 stored in transaction response buffer 205 which is currently being serviced and sends current response tag 211 to tag comparator 250.
In the above example the tags generated by tag generator circuit 230 represented all the bits in each address of transaction response buffer 205. In other embodiments, the tags may only be some number of significant bits of the available addresses. The number of most significant bits (MSBs) used determines the granularity with which the individual locations can be addressed. For example, if transaction response buffer 205 contains 64K addressable locations, then this corresponds to #0000h through #FFFFh addresses. If only the 3 MSBs were used as a tag (e.g. #FFFh), then the addresses will only be accessible in blocks of 16 bytes. This would change how tag generator circuit 230 generates each next tag 231. When next tag 231 is generated, the memory request size would be added to the tag stored in tag register 240 then, if next tag 231 did not fall on an addressable block boundary, next tag 231 would be rounded up to the next higher block size. For example, assuming the 64K address space described above and using 3 MSB. If the first memory request is 4 bytes, the first tag is #000h and next tag 231 would be #001h. Since #0000h+#0004h=#0004h is not addressable on a block boundary, then #0004h gets rounded up to the next addressable block, which is #001h. When another memory request 226 of 16 bytes is received by transaction request generator circuit 220, memory request 226 is encoded with the tag held in tag register 240, which is #001h. Then tag generator circuit 230 generates next tag 231 by adding the 16 bytes to the tag held in tag register 240. In this example, next tag 231 would be #002h, since #002h does fall on an addressable block boundary. Next tag 231 is stored in tag register 240 awaiting the next memory request.
Numerous variations and modifications will become apparent to those skilled in the art once the above disclosure is fully appreciated. It is intended that the following claims be interpreted to embrace all such variations and modifications.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5197135 *||Jun 26, 1990||Mar 23, 1993||International Business Machines Corporation||Memory management for scalable compound instruction set machines with in-memory compounding|
|US5208914||Jun 11, 1990||May 4, 1993||Superconductor Systems Limited Partnership||Method and apparatus for non-sequential resource access|
|US5325487||Aug 14, 1990||Jun 28, 1994||Integrated Device Technology, Inc.||Shadow pipeline architecture in FIFO buffer|
|US5915110 *||Nov 20, 1997||Jun 22, 1999||Advanced Micro Devices, Inc.||Branch misprediction recovery in a reorder buffer having a future file|
|US5996036||Jan 7, 1997||Nov 30, 1999||Apple Computers, Inc.||Bus transaction reordering in a computer system having unordered slaves|
|US6292807 *||Dec 15, 1998||Sep 18, 2001||Micron Technology, Inc.||Method for controlling pipelined memory access requests|
|EP0520835A2||Jun 29, 1992||Dec 30, 1992||Digital Equipment Corporation||Scheme for interlocking devices in a computer system|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7047374 *||Feb 5, 2003||May 16, 2006||Intel Corporation||Memory read/write reordering|
|US7263577 *||Mar 3, 2005||Aug 28, 2007||Qualcomm Incorporated||Power saving methods and apparatus to selectively enable comparators in a CAM renaming register file based on known processor state|
|US7636769||Apr 14, 2006||Dec 22, 2009||Microsoft Corporation||Managing network response buffering behavior|
|US8244950 *||Nov 1, 2002||Aug 14, 2012||Advanced Micro Devices, Inc.||Buffering non-posted read commands and responses|
|US8316194||Dec 15, 2009||Nov 20, 2012||Intel Corporation||Mechanisms to accelerate transactions using buffered stores|
|US8365016||Nov 30, 2011||Jan 29, 2013||Intel Corporation||Performing mode switching in an unbounded transactional memory (UTM) system|
|US8521995||Dec 15, 2009||Aug 27, 2013||Intel Corporation||Handling operating system (OS) transitions in an unbounded transactional memory (UTM) mode|
|US8677045 *||Sep 23, 2011||Mar 18, 2014||Stmicroelectronics (Grenoble 2) Sas||Transaction reordering system and method with protocol indifference|
|US8856466||Oct 23, 2012||Oct 7, 2014||Intel Corporation||Mechanisms to accelerate transactions using buffered stores|
|US8886894||Oct 23, 2012||Nov 11, 2014||Intel Corporation||Mechanisms to accelerate transactions using buffered stores|
|US8990436||May 29, 2013||Mar 24, 2015||Stmicroelectronics S.R.L.||Method for handling access transactions and related system|
|US9021482 *||May 4, 2007||Apr 28, 2015||International Business Machines Corporation||Reordering data responses using ordered indicia in a linked list|
|US9069670||Oct 23, 2012||Jun 30, 2015||Intel Corporation||Mechanisms to accelerate transactions using buffered stores|
|US9195600||Nov 10, 2014||Nov 24, 2015||Intel Corporation||Mechanisms to accelerate transactions using buffered stores|
|US20030177320 *||Feb 5, 2003||Sep 18, 2003||Suneeta Sah||Memory read/write reordering|
|US20040024947 *||Nov 1, 2002||Feb 5, 2004||Frank Barth||Buffering non-posted read commands and responses|
|US20040024948 *||Nov 1, 2002||Feb 5, 2004||Joerg Winkler||Response reordering mechanism|
|US20060206688 *||Mar 3, 2005||Sep 14, 2006||Bridges Jeffrey T||Power saving methods and apparatus to selectively enable comparators in a CAM renaming register file based on known processor state|
|US20070112579 *||Aug 31, 2006||May 17, 2007||Ads Alliance Data Systems, Inc.||Market management system|
|US20070244993 *||Apr 14, 2006||Oct 18, 2007||Microsoft Corporation||Managing network response buffering behavior|
|US20080276240 *||May 4, 2007||Nov 6, 2008||Brinda Ganesh||Reordering Data Responses|
|US20110145512 *||Jun 16, 2011||Ali-Reza Adl-Tabatabai||Mechanisms To Accelerate Transactions Using Buffered Stores|
|US20110145552 *||Dec 15, 2009||Jun 16, 2011||Koichi Yamada||Handling Operating System (OS) Transitions In An Unbounded Transactional Memory (UTM) Mode|
|US20110145637 *||Dec 15, 2009||Jun 16, 2011||Jan Gray||Performing Mode Switching In An Unbounded Transactional Memory (UTM) System|
|US20110208921 *||Feb 19, 2010||Aug 25, 2011||Pohlack Martin T||Inverted default semantics for in-speculative-region memory accesses|
|US20120079148 *||Sep 29, 2011||Mar 29, 2012||Stmicroelectronics S.R.L.||Reordering arrangement|
|WO2011081718A3 *||Nov 10, 2010||Aug 25, 2011||Intel Corporation||Mechanisms to accelerate transactions using buffered stores|
|U.S. Classification||712/245, 711/220, 710/112, 718/106, 710/40, 718/103, 711/145, 712/215, 711/156|
|Apr 11, 2000||AS||Assignment|
Owner name: ADVANCED MICRO DEVICES, INC., CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:MIRANDA, PAUL C.;HEWITT, LARRY D.;ENNIS, STEPHEN C.;REEL/FRAME:010707/0376
Effective date: 20000405
|Sep 26, 2006||FPAY||Fee payment|
Year of fee payment: 4
|Aug 18, 2009||AS||Assignment|
Owner name: GLOBALFOUNDRIES INC., CAYMAN ISLANDS
Free format text: AFFIRMATION OF PATENT ASSIGNMENT;ASSIGNOR:ADVANCED MICRO DEVICES, INC.;REEL/FRAME:023119/0083
Effective date: 20090630
|Oct 25, 2010||FPAY||Fee payment|
Year of fee payment: 8
|Oct 29, 2014||FPAY||Fee payment|
Year of fee payment: 12